Skip to main content

Lovable Apps Leak User Data: 24,664 Sites Served Records Without Login

Hamburg security firm Crissy Field, working with Handelsblatt, analysed 159,041 apps built on the AI app builder Lovable. Of 34,545 that used Supabase, 24,664 sent database records to visitors without a login, and 7,403 exposed tables containing email addresses. Exposed data included passport numbers, 200,000 forum chat messages, and ticket payment records. The cause was misconfigured database access, not a single Lovable bug.

On this page

What the Lovable scan found

Lovable is a Swedish startup whose AI turns plain-language prompts into working web apps. Crissy Field, a security firm in Hamburg, set out to measure how safely those apps handle data. Using the open Common Crawl web archive, it identified 159,041 Lovable projects published between December 2025 and September 2026, and found that 34,545 of them used Supabase as their database.

Of those Supabase apps, 24,664 sent database records to the browser as soon as the page loaded, with no login or other interaction. The researchers note that this alone is not a flaw, because some data is meant to be public. So they made further read-only requests with the same key every visitor's browser receives, and found records containing email addresses in 10,560 tables across 7,403 sites, about 30% of the sites examined closely.

The firm cautions that the 30% figure is not a confirmed leak rate: some addresses may be public contact details or test data, and a Common Crawl sample is not representative of every Lovable app. The work was done with the German business paper Handelsblatt, which first reported it on 2 November; the English report carries updated figures.

Passports, chats, and payment records

The examples are what make the numbers concrete. A German lifestyle forum exposed more than 80,000 profiles, with full names, email addresses, dates of birth, and gender, plus more than 200,000 chat messages. Event organisers and restaurants exposed tens of thousands of ticket and food sales records, including phone numbers, postal addresses, payment information, and ticket codes.

One service provider left more than 1,000 identity document records readable, including passport numbers, expiry dates, and nationalities. A property manager exposed AI-generated summaries of internal emails about tenants and owners, and one provider's 280,000 customer emails were accessible. The researchers say they could not tell whether any of it had been misused, and they are not publishing how to reach the data.

Why the database key was enough

Supabase apps talk to the database straight from the browser using a key that ships inside the page. That design is only safe when the database itself decides who may read each row. Supabase's documentation is blunt about the default risk: a table in an exposed schema without row level security "is readable and writable by any role with a grant on it". Once row level security is on, no data flows through the public key until the developer writes policies.

Crissy Field describes the problem as recurring access control mistakes rather than one bug, and warns that because Lovable apps wire up their databases in similar ways, a technique that works on one app could work on many others with little extra effort. The study did not examine Lovable's own built-in database features, added in late 2025. According to the researchers, Lovable's terms place responsibility for data handling on its customers.

What builders and users should take from it

AI app builders remove the hard part of shipping software, and with it the moment when a developer would normally stop and think about permissions. An app can look finished and still hand its whole user table to any visitor. Crissy Field's advice is to have access permissions, interfaces, and separation between user accounts reviewed before an app handles real customer data, and to repeat that review after significant changes.

If you built an app this way, open each table in Supabase, confirm row level security is enabled, and test reading data while logged out. If you are a user, treat a small, newly launched app the way you would any unknown service: share as little personal data as it needs, and never upload identity documents unless you trust who is holding them.

CuriousLM runs supported AI models locally on your device. Try CuriousLM.